Predictability in Scale-specific Systems

This paper investigates the hump-shaped behavior of slopes and coefficients of determination of predictive regressions for future excess market returns on past regressors, as discussed by Bandi, Perron, Tamoni, and Tebaldi (2019). In doing so, the results of the former paper for twoway aggregated regression models and classical predictive systems are first reviewed, after which the scale-specific framework proposed by Bandi et al. (2019) is extended threefold. Firstly, a robustness analysis is conducted to investigate the sensitivity of presented results to slight model changes. Secondly, a high-frequency analysis is employed. Thirdly, this research formally addresses the idea of incorporating multiple regressors in scale-wise predictive systems. This paper finds that hump-shaped behavior and scale-specific predictability remain to hold under altered data and sample adjustments, while the location of predictability peaks may vary over scales. High-frequency data supports the findings of Bandi et al. (2019) for the NYSE/AMEX and S&P 500 indices, while the occurrence of hump-shaped behavior in other markets is not definitive. Adding multiple regressors preserves hump-shaped behavior, while predictability reaches up to 90%.